SocialReasonBench - Counterfactual Reasoning: leaderboard
Metric: Accuracy (%; the 50 counterfactual reasoning questions of SocialReasonBench; six-option multiple-choice questions on counterfactual narrative gameplay clips of Detroit: Become Human, labels grounded in the game's script, flowchart and recorded branches; zero-shot, temperature 0 or 0.01; Gemini, Qwen3-Omni, MiniCPM-o and ARC-Hunyuan-Video read native video with audio, GPT-4o, Claude 3.5 Sonnet and GLM-4.6V-Flash 32 evenly sampled keyframes without audio). Source: arxiv.org. Saturation forecast: Around October 2027. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro | 62 |
| 2 | Claude 3.5 Sonnet (20240620) | 56 |
| 3 | GPT-4o (2024-08-06) | 50 |
| 4 | Gemini 3 Flash | 46 |
| 5 | Qwen3 Omni 30B A3B Instruct | 36 |
Interactive version: theaggregate.ai/benchmark?slug=socialreasonbench-counterfactual-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-26.